The agency of record was an answer to two shortages. Making an advertisement was expensive and slow, so the client needed a house full of people who could make one. Buying media meant telephoning a sales representative at a newspaper, a network or a billboard company, so the client needed a house that already knew them. In exchange, the client gave one agency the whole of its advertising, for years, and paid it a share of what it spent. The arrangement made sense for as long as its two premises held.
They no longer hold. Production is abundant: a written brief can become a finished film, in every crop a media plan needs, in an afternoon. Media is bought by machines through auctions that no relationship can influence. And the asset that decides whether advertising works, the first-party data about who bought what and why, sits inside the client's own systems, where an agency can only borrow it. A chief marketing officer who signs a new agency-of-record contract in these conditions is renting a factory for a product that is now made on a laptop, and paying rent on a warehouse the landlord is allowed to keep.
What replaces it is not the absence of agencies. It is an advertising system the client owns, in its own accounts and on its own contracts, and a different kind of firm: one paid to build the system, transfer it and then advise, rather than to run it for as long as the retainer lasts. This article sets out the argument, the shape of the owned system, the path from here to there, and the ways of getting it wrong.
What the agency of record was for
For most of the last century, an advertisement was a piece of industrial production. A television commercial needed a director, a crew, a studio, film stock, an edit suite and weeks. A print campaign needed photographers, typesetters and a production department. No client outside the very largest could keep that capacity in the building for the months between campaigns, so the agency held it and spread the cost across clients. Creative talent clustered in agencies for the same reason: it was where the work was.
Media was bought by hand. Inventory was sold by people, priced by negotiation, and allocated on the strength of the relationship and the size of the order. An agency that bought for many clients bought at better prices and knew what was available. It also held the contracts, the invoices and the discounts, and it was paid by commission on the media it placed, later supplemented and then replaced by retainers priced on the staff assigned to the account.
The model concentrated everything in the agency's hands: the ideas, the assets, the media contracts, the audience knowledge and the institutional memory of what had worked. The client kept the brand and the budget. That concentration was the point. It was efficient, and it was the reason an agency-of-record relationship, once made, was so hard to leave.
Both conditions have ended
Production is no longer scarce
Generative production has changed the cost of a finished asset from a studio budget to a licence fee. Concepts, scripts, still and moving images, voice, edits and the dozens of variants a media plan demands can be produced by a small team with the right tooling, reviewed against a brand standard and released through a governed approval path. Large shoots still exist and still matter for the work that deserves them, but they are a choice rather than a bottleneck. The capacity an agency held because nobody else could afford it is now a capability a client can install. The EU AI Act places disclosure obligations on some synthetic content, and a client that owns its production line can meet them in the tooling; a client that rents its production cannot see whether they were met.
Media is bought by machines
Search, social, video, retail media and most display inventory are bought through auctions in real time, in accounts that belong to whoever opened them. Price is set by the auction, not by the sales lunch. The buying power that came from aggregating clients has shrunk to the channels where inventory is still sold by people: some television, some out-of-home, sponsorships and partnerships. Those remain, and they are a fraction of what they were.
The decisive asset moved inside the client
The data that makes advertising work is first-party: who bought, who renewed, who left, who called the service line and what for. It lives in the client's commerce, service and finance systems. Privacy law, including the GDPR, the revised Swiss FADP, Jordan's Personal Data Protection Law and the US state privacy laws, constrains how that data may be collected, shared and used, and places the obligation on the company whose customers they are. Third-party identifiers are less available than they were and less reliable where they remain. An agency can be granted access to first-party data; it cannot own it, and it should not be the party that decides how it is used. The centre of gravity moved, and the contract has not followed it.
The incentive problem
Commission on media rewards spending. Every plan the agency writes is written by a party that earns more when the plan is larger, and the client is asked to trust that the plan is nevertheless the right size. Retainers priced on assigned staff reward a large team and a complex account, and penalise the agency for making the work simpler. Neither arrangement is dishonest. Both are structures in which the party advising on a decision is paid in proportion to one of its outcomes, and no audit committee would accept that structure from an auditor or a procurement adviser.
The structure also produces opacity as a by-product. Where an agency buys inventory as principal and resells it, or accepts rebates from platforms and vendors, the client's cost of media is not the price the market charged. Where the agency's own accounts hold the campaigns, the client cannot see the price at all. Where the agency's strategists brief the agency's producers and the agency's planners measure the result, the report the board receives has been written by the party it is about.
The party advising on the size of the budget should not be paid in proportion to the size of the budget. Every other function in the company already lives by that rule.
An advertising system the client owns
An owned system has four parts, and ownership means something specific for each: the asset is in the client's name, on the client's contract, in the client's infrastructure, and it keeps working when any supplier is removed.
| Component | Under an agency of record | In an owned system |
|---|---|---|
| Data | Audiences and pixels in agency accounts; client data shared under agency terms | Consent, tagging, identity and audiences in the client's warehouse and platforms, with a register the privacy function owns |
| Creative system | Assets made and stored by the agency; the standard lives in people's heads | A written brand and creative standard, a governed library, a production line the client operates, an approval record on every asset |
| Measurement | Platform-reported returns forwarded in a monthly deck | Marketing-mix modelling and incrementality experiments in the client's warehouse, reconciled with finance, registered before they run |
| Media procurement | Contracts, discounts and rebates held by the agency | Accounts opened in the client's name; media bought at platform cost; every fee on its own invoice line |
The data layer is the foundation. Consent management, server-side tagging, first-party identity and audience construction are built once, with the lawful basis for every flow recorded, and every other component reads from them. The creative system is a production line rather than a department: a standard that says what may be shown, said and claimed per market; a library with a version history; an approval path with named roles; and machine-assisted production behind it, disciplined by a per-job budget and by the same review a hand-made asset would receive. Measurement answers the finance question, which is not what the platform attributed but what the spending caused, in units the board already uses, and it is built by people who do not also spend the budget. Media is bought in the client's accounts, at the price the platform charges, by people whose fee does not move with the total.
What a firm should be paid to do
If the client owns the system, the firm's role changes and so should its fee. There are three things worth paying for, in sequence.
Build. Design the account architecture, the data-flow register, the measurement model, the creative standard and the production line; stand them up in the client's name; run the first campaigns and the first experiments at controlled budgets. This is project work with deliverables, gates and a fixed or capped price, and each deliverable is assigned to the client on payment.
Transfer. Operate the system with the client's team taking the controls one at a time, from a runbook that names every recurring task: how to plan a quarter, approve an asset, launch a campaign, run a test and answer a data-subject request. The engagement is designed to shrink. A firm that resists the shrinking has told you what it is optimising for.
Advise. Once the system is the client's, the firm's value is the outside view: the quarterly review of what the model says and what the experiments showed, the redesign when a channel or a regulation changes, the second opinion on a large creative investment, the audit of the controls. This is a retainer, but a small one, priced on the advice and not on the media or the headcount.
None of these fees moves with the size of the budget. That is the whole point.
The transition for a chief marketing officer
The path from an agency of record to an owned system has three stages, and the order matters.
Audit first. List every account, tag container, pixel, dataset, contract and licence that the company's advertising depends on, and record whose name is on each. Read the media contracts for principal buying, rebates and the ownership of data. Reconstruct what the last year's spending can be shown to have caused, as opposed to what the platforms attributed to it. The result is usually uncomfortable and always useful. It is the baseline against which the new system will be judged, and it tells you how dependent you are.
Decouple next. Separate the functions that the agency-of-record contract bundled: creative, media, data and measurement. Move accounts and contracts into the company's name, or open new ones and migrate. End arrangements in which the same party plans, spends and measures. Insist that every fee appear on its own line. This stage can be done with the incumbent agency, and a good one will help; the resistance you meet is information.
In-house what is strategic, and only that. The data layer, the measurement function, the account architecture, the brand and creative standard, and the annual and quarterly planning are strategic; they encode what the company knows about its customers and what it is willing to say to them, and they should be owned by employees. Large productions, specialist channels, the occasional big idea and surge capacity are episodic; buy them, from the best supplier for each, on project terms. The common mistake is to in-house execution and outsource judgement. It should be the reverse.
What agencies still do well, and how in-housing fails
An honest argument acknowledges what the old model did that an owned system does not do by itself.
Agencies concentrate creative talent and give it a career, and the best of them produce ideas that a marketing department, living inside the company's own assumptions, rarely does. They see across categories and markets, and they bring the outside view that an in-house team loses over time. They manage large productions well, because they do it constantly. They still buy the hand-sold channels better than a client can. And they take a kind of risk with an idea that an employee, whose next review depends on this quarter, is right to avoid. None of these needs an agency of record. Each can be bought when it is needed, by a client who owns the system into which the work is delivered.
In-housing fails in predictable ways. The first is hiring for execution: a team of buyers and producers with no strategist, which becomes an order-taker for the business units and a customer for the platforms' sales teams, whose advice arrives free and is not neutral. The second is measurement built by the people it measures, which reproduces the agency's conflict inside the company. The third is a stale team: in-house creative that has seen one brand for years and nothing else. The fourth is tooling debt: a system assembled from whatever each hire already knew, with no account architecture, no register and no runbook, which is as hard to leave as the agency was. The fifth is cost moved rather than removed, where the retainer becomes a headcount line and the board asks, fairly, what changed. Each of these is avoided by the same discipline: build the system before you hire the team, write the runbook before you take the controls, and keep a firm outside the building whose fee does not depend on how much you spend.
What to do on Monday
- Ask for the list of every advertising account, container, pixel, dataset and contract the company depends on, with the name on each. If the list cannot be produced within the week, that is the first finding.
- Read the media agreement for principal buying, rebates, data ownership and exit terms, with legal and procurement in the room.
- Ask finance what the last year of advertising can be shown to have caused, and compare the answer with the platform-reported figures in the last agency deck.
- Decide which functions are strategic to your company and will be owned by employees, and which are episodic and will be bought on project terms.
- Commission the audit as a fixed-price piece of work from a firm that does not buy media, and write into the brief that every deliverable is yours on payment.
- Set the date by which every account and contract is in the company's name, and tell the incumbent agency, so that the transition is a plan rather than a surprise.